Evidence Instantiation (5.0.4)

This new tool creates a network based on the current network and a set of evidence. While the original network and its evidence define a subspace in the original Joint Probability Distribution (JPD), the new network is entirely dedicated to the representation of that subspace.

The structure of the new network does not change. The only modifications are in Conditional Probability Distributions.

Note that the new graph cannot always exactly represent the subspace described by the set of evidence. Indeed, some conditional dependencies can be introduced in the original network through the Likelihood set in the collider nodes of V structures. Those conditional dependencies are not representable in the instantiated graph without modifying the structure.

For example, [Factor_5] and [Factor_6] are collider nodes. Setting evidence on those nodes will prevent the exact representation of the defined subspace. As illustrated in the monitors shown in the previous slide, the approximation is propagated to [Factor_0], [Factor_3] and [Factor_7].

This feature can be used for setting evidence on nodes while being able to keep them in the set of “drivers” (e.g. Target Dynamic Profile) or analyzed nodes (e.g. Target Analysis Report).